Memory error, and ProgressMeter strangeness during HaplotypeCaller
Posted in Ask the GATK team | Last updated on 2014-04-17 18:58:09

Comments (23)

Log attached. It's become hard to tell what the columns mean in the ProgressMeter output, but I know this is strange. I see six columns in the output, but the header line has apparently nine columns space-delimited. The second column is all zeros, and fourth approaches infinity before my memory error. "because you did not provide enough memory"

The crash is probably more pertinent. This morning when I found it had crashed, I saw that I had not specified any -Xmx argument to java, so I changed it to the amount of RAM (32G) divided by the -nct parameter 8, for about 3700MB each. That's a worst case scenario of duplicating ram usage per thread, so now I'm stumped why and how is 3.7GB not enough for a read-walker ??

For now I'm going to try -nct 4 and -Xmx7G

edit: Re-trying with -nct 4 and -Xmx7G seems to show the same behavior; I have zero activeregions, and a hundreds of weeks in column four. My estimated runtime is fluctuating between 70 hours and 6 days; so I expect the same memory crashout in a few hours. Will update then.

Return to top Comment on this article in the forum